The appropriate use of serum creatinine level as a surrogate for time in the course of renal failure when dialysis commences requires it to be a significant predictor of mortality in incident patients with end-stage renal disease (ESRD). This study evaluated factors that account for variations in creatinine level before the initiation of dialysis and whether incident creatinine level after controlling for these factors was a risk factor for mortality. This is a retrospective cohort study of patients from Maryland and Virginia who initiated dialysis between April 1, 1995, and December 31, 1996, with data ascertained from the Health Care Financing Administration Form 2728. Multivariate models were used to evaluate both the factors that predict incident serum creatinine level and the association between creatinine level and mortality. There were 5, 388 patients followed up for an average of 23.6 +/- 0.2 months. Mean creatinine level was 9.2 +/- 0.1 mg/dL, with case-mix factors most predictive of serum creatinine level and accounting for 9% of its variance. Hematocrit and blood urea nitrogen levels as additional surrogates for progression of renal disease accounted for 7.4% of the variance, whereas the nutritional parameters, body mass index, and albumin level only explained an additional 1% of the total variance in creatinine level. Creatinine level was inversely correlated with mortality risk, and this relationship was sustained both with transformation into an estimated glomerular filtration rate and multivariate adjustment for confounders (relative risk = 0. 96; P < 0.0001). Creatinine values from an incident ESRD population have a weak relationship with the timing of dialysis initiation but represent a strong measure of health status.